Files
borgmatic/borgmatic/actions/pattern.py
T

331 lines
14 KiB
Python

import glob
import itertools
import logging
import os
import pathlib
import borgmatic.borg.pattern
logger = logging.getLogger(__name__)
def parse_pattern(pattern_line, default_style=borgmatic.borg.pattern.Pattern_style.NONE):
'''
Given a Borg pattern as a string, parse it into a borgmatic.borg.pattern.Pattern instance and
return it.
'''
try:
(pattern_type, remainder) = pattern_line.split(' ', maxsplit=1)
except ValueError:
raise ValueError(f'Invalid pattern: {pattern_line}')
try:
(parsed_pattern_style, path) = remainder.split(':', maxsplit=1)
pattern_style = borgmatic.borg.pattern.Pattern_style(parsed_pattern_style)
except ValueError:
pattern_style = default_style
path = remainder
return borgmatic.borg.pattern.Pattern(
path,
borgmatic.borg.pattern.Pattern_type(pattern_type),
borgmatic.borg.pattern.Pattern_style(pattern_style),
source=borgmatic.borg.pattern.Pattern_source.CONFIG,
)
def collect_patterns(config, working_directory):
'''
Given a configuration dict and the working directory, produce a single sequence of patterns
comprised of the configured source directories, patterns, excludes, pattern files, and exclude
files.
The idea is that Borg has all these different ways of specifying includes, excludes, source
directories, etc., but we'd like to collapse them all down to one common format (patterns) for
ease of manipulation within borgmatic.
'''
try:
return (
tuple(
borgmatic.borg.pattern.Pattern(
source_directory,
source=borgmatic.borg.pattern.Pattern_source.CONFIG,
)
for source_directory in config.get('source_directories', ())
)
+ tuple(
parse_pattern(pattern_line.strip())
for pattern_line in config.get('patterns', ())
if not pattern_line.lstrip().startswith('#')
if pattern_line.strip()
)
+ tuple(
parse_pattern(
f'{borgmatic.borg.pattern.Pattern_type.NO_RECURSE.value} {exclude_line.strip()}',
borgmatic.borg.pattern.Pattern_style.FNMATCH,
)
for exclude_line in config.get('exclude_patterns', ())
)
+ tuple(
parse_pattern(pattern_line.strip())
for filename in config.get('patterns_from', ())
for expanded_path in expand_directory(filename, working_directory)
for pattern_line in open(expanded_path, encoding='utf-8')
if not pattern_line.lstrip().startswith('#')
if pattern_line.strip()
)
+ tuple(
parse_pattern(
f'{borgmatic.borg.pattern.Pattern_type.NO_RECURSE.value} {exclude_line.strip()}',
borgmatic.borg.pattern.Pattern_style.FNMATCH,
)
for filename in config.get('exclude_from', ())
for expanded_path in expand_directory(filename, working_directory)
for exclude_line in open(expanded_path, encoding='utf-8')
if not exclude_line.lstrip().startswith('#')
if exclude_line.strip()
)
)
except (FileNotFoundError, OSError) as error:
logger.debug(error)
raise ValueError(f'Cannot read patterns_from/exclude_from file: {error.filename}')
def expand_directory(directory, working_directory):
'''
Given a directory path, expand any tilde (representing a user's home directory) and any globs
therein. Return a list of one or more resulting paths.
Take into account the given working directory so that relative paths are supported.
'''
expanded_directory = os.path.expanduser(directory)
# This would be a lot easier to do with glob(..., root_dir=working_directory), but root_dir is
# only available in Python 3.10+.
normalized_directory = os.path.join(working_directory or '', expanded_directory)
glob_paths = glob.glob(normalized_directory)
if not glob_paths:
return [expanded_directory]
working_directory_prefix = os.path.join(working_directory or '', '')
return [
(
glob_path
# If these are equal, that means we didn't add any working directory prefix above.
if normalized_directory == expanded_directory
# Remove the working directory prefix added above in order to make glob() work. We
# can't use os.path.relpath() here because it collapses any use of Borg's slashdot hack.
else glob_path.removeprefix(working_directory_prefix)
)
for glob_path in glob_paths
]
def expand_patterns(patterns, working_directory=None, skip_paths=None):
'''
Given a sequence of borgmatic.borg.pattern.Pattern instances and an optional working directory,
expand tildes and globs in each root pattern and expand just tildes in each non-root pattern.
The idea is that non-root patterns may be regular expressions or other pattern styles containing
"*" that borgmatic should not expand as a shell glob.
Return all the resulting patterns as a tuple.
If a set of paths are given to skip, then don't expand any patterns matching them.
'''
if patterns is None:
return ()
return tuple(
itertools.chain.from_iterable(
(
(
borgmatic.borg.pattern.Pattern(
expanded_path,
pattern.type,
pattern.style,
pattern.device,
pattern.source,
)
for expanded_path in expand_directory(pattern.path, working_directory)
)
if pattern.type == borgmatic.borg.pattern.Pattern_type.ROOT
and pattern.path not in (skip_paths or ())
else (
borgmatic.borg.pattern.Pattern(
os.path.expanduser(pattern.path),
pattern.type,
pattern.style,
pattern.device,
pattern.source,
),
)
)
for pattern in patterns
),
)
def get_existent_path_or_parent(path):
'''
Given a path, return it if it exists. Otherwise, return the longest parent directory of the path
that exists. Return None if none of these paths exist.
This is used below for finding an existent path prefix of pattern's path, which is necessary if
the path contain globs or other special characters that we don't want to try to interpret
(because we want to leave that responsibility to Borg).
'''
if path.startswith('/e2e/'):
return None
try:
return next(
candidate_path
for candidate_path in (
path,
*tuple(str(parent) for parent in pathlib.PurePath(path).parents),
)
if os.path.exists(candidate_path)
)
except StopIteration:
return None
def device_map_patterns(patterns, working_directory=None):
'''
Given a sequence of borgmatic.borg.pattern.Pattern instances and an optional working directory,
determine the identifier for the device on which the pattern's path resides—or None if the path
doesn't exist or is from a non-root pattern. Return an updated sequence of patterns with the
device field populated. But if the device field is already set, don't bother setting it again.
This is handy for determining whether two different pattern paths are on the same filesystem
(have the same device identifier).
This function only considers the start of a pattern's path—from the start of the path up until
there's a path component with a glob or other non-literal character. If there are no such
characters, the whole path is considered. The rationale is that it's not feasible for borgmatic
to interpret Borg's patterns to see which actual files (and therefore devices) they map to. So
for instance, a pattern with a path of "/var/log/*/data" would end up with its device set to the
device of "/var/log"—ignoring the "/*/data" part due to that glob.
The one exception is that if a regular expression pattern path starts with "^", that will get
stripped off for purposes of determining its device.
'''
return tuple(
borgmatic.borg.pattern.Pattern(
pattern.path,
pattern.type,
pattern.style,
device=pattern.device or (os.stat(existent_path).st_dev if existent_path else None),
source=pattern.source,
)
for pattern in patterns
for existent_path in (
get_existent_path_or_parent(
os.path.join(working_directory or '', pattern.path.lstrip('^')),
),
)
)
def deduplicate_runtime_directory_patterns(patterns, config, borgmatic_runtime_directory=None):
'''
Given a sequence of borgmatic.borg.pattern.Pattern instances, the borgmatic runtime directory,
and a configuration dict, return them without any duplicate root child patterns that contain the
runtime directory. For instance, if two root patterns are given with paths "/foo" and
"/foo/bar", and the runtime directory is "/foo/bar", return just the "/foo" pattern. Non-root
patterns and patterns not containing the runtime directory are passed through without
modification.
One exception to deduplication is if two paths are on different filesystems (devices) and
"one_file_system" is True in the given configuration. In that case, the paths won't get
deduplicated, because Borg won't cross filesystem boundaries when "one_file_system" is True.
The idea is that if Borg is given a root parent pattern containing the borgmatic runtime
directory, then Borg doesn't also need to be given child patterns, because it will naturally
spider the contents of the parent pattern's path. Additionally, there are cases where Borg
coming across the same file twice will result in duplicate reads and even hangs, e.g. when a
database hook in the borgmatic runtime directory is using a named pipe for streaming database
dumps to Borg.
This deduplication is limited to the borgmatic runtime directory (where borgmatic's named pipes
exist), because there are other legitimate use cases for parent and child patterns to both exist
in patterns. For instance, with some snapshotted filesystems, snapshots don't traverse from a
parent filesystem to a child and therefore both need to remain in patterns.
And for the case of named pipes outside of the borgmatic runtime directory, there is code
elsewhere (in the "create" action) that auto-excludes special files to prevent Borg hangs.
'''
if borgmatic_runtime_directory is None:
return patterns
deduplicated = {} # Use just the keys as an ordered set.
runtime_directory_parents = set(pathlib.PurePath(borgmatic_runtime_directory).parents).union(
{pathlib.PurePath(borgmatic_runtime_directory)}
)
for pattern in patterns:
if pattern.type != borgmatic.borg.pattern.Pattern_type.ROOT:
deduplicated[pattern] = True
continue
pattern_parents = pathlib.PurePath(pattern.path).parents
# If:
#
# 1. another pattern is a parent of the current pattern (even n levels up),
# 2. both patterns are parents of the runtime directory (even n levels up),
# 3. and both patterns are on the same filesystem (or one_file_system is not set)
#
# ... then consider the current pattern as a duplicate.
for other_pattern in patterns:
if other_pattern.type != borgmatic.borg.pattern.Pattern_type.ROOT:
continue
device_matches = pattern.device is not None and (
other_pattern.device == pattern.device or config.get('one_file_system') is not True
)
if (
pathlib.PurePath(other_pattern.path) in pattern_parents
and pathlib.PurePosixPath(other_pattern.path) in runtime_directory_parents
and pathlib.PurePosixPath(pattern.path) in runtime_directory_parents
and device_matches
):
break
else:
deduplicated[pattern] = True
return tuple(deduplicated.keys())
def process_patterns(
patterns, config, working_directory, borgmatic_runtime_directory=None, skip_expand_paths=None
):
'''
Given a sequence of Borg patterns, a configuration dict, a configured working directory, the
borgmatic runtime directory, and a sequence of paths to skip path expansion for, expand and
deduplicate any "root" patterns, returning the resulting root and non-root patterns as a list.
If the borgmatic runtime directory is None, then don't deduplicate patterns. Deduplication is
really only necessary for the "create" action when the runtime directory might contain named
pipes for database dumps.
'''
skip_paths = set(skip_expand_paths or ())
return list(
deduplicate_runtime_directory_patterns(
device_map_patterns(
expand_patterns(
patterns,
working_directory=working_directory,
skip_paths=skip_paths,
),
),
config,
borgmatic_runtime_directory,
),
)